Prompt Injection on SQuAD Inj benchmark leaderboard · SOTA2 Research

Prompt Injection on SQuAD Inj

1.11ASR (Naive)Robustness via Referencing

Evaluation Results

Method	Links
Robustness via Referencing2025.04		1.11	0.22	0.56	0.78	0.11
Robustness via Referencing2025.04		2.22	1.44	1.22	0.44	1.22
Robustness via Referencing2025.04		2.78	0.22	2	0.78	0
Sandwich2025.04		8.11	24.44	8.22	9.44	33.22
Sandwich2025.04		9.78	15.78	7.78	3.78	13.11
Sandwich2025.04		10.11	32.22	8	48.33	46.33
Spotlight2025.04		15.44	57.89	14.44	77.67	85.67
Reminder2025.04		19.33	32.11	19.56	22.67	42.78
Spotlight2025.04		22.56	32.78	19.33	79.22	81.33
None2025.04		22.67	72.67	26.33	60	80.78
Instructional2025.04		23.56	37.44	23.89	34.11	42.56
Spotlight2025.04		26	67.89	29.11	97.56	99.11
None2025.04		35.33	82.44	31.56	74.89	91.67
Instructional2025.04		42.33	46.89	42.56	91.22	75.44
Reminder2025.04		42.67	71.89	40.11	38	69.78
None2025.04		44.78	91.67	50.33	98.22	96.67
Reminder2025.04		46.78	71.56	46.44	87.78	69.44
Instructional2025.04		47.67	70.78	42.78	50.56	79.33